Saymock

Collaborative design board for AI agents: your agent designs screens as editable HTML + Tailwind while your team watches, comments and edits live.

Serveur MCP hébergé

npx add-mcp 'https://app.saymock.com/mcp'

S’installe dans Claude Code, Codex, Cursor et plus

Documentation

Your agent, your account: Saymock never calls an AI model itself. Plug in Claude Code or any MCP client.

One sentence in, screens out. Then refine them together.

  1. 01

    Connect your agent

    One command adds Saymock to Claude Code or any MCP client. The agent acts with your permissions, in your workspace, and can read your code.
  2. 02

    Describe the screen

    The agent builds frames in HTML + Tailwind, checks a screenshot of the result and fixes layout issues before handing it back to you.
  3. 03

    Refine together

    Edit by hand while it works. Pin a comment that mentions @claude: it picks up the element, checks its work and replies in the thread.

Your agents, on the board

They show up in presence like any teammate, in their own color. Their edits are highlighted and undo separately from yours, and every request becomes a task tracked in its thread.

Layers ⌥1

Week

Hi, Léa

Today’s workout

Schedule

Start buttonClaude

Workout in progress

Summary

iPhone 15 · Week

9:41100%

Tuesday, November 5

Hi, Léa

Today’s workout Easy run, 5 km Pace 6:10 /km · 31 min

  • Wed Rest
  • Thu Intervals 8 × 400 m
  • Sat Long run 14 km

Claude editing

Tristan

T Tristan · 2 min ago

@claude make the button full width and check it at 390 px.

Claude · just now

Done: added w-full, checked the layout at 390 px, no overflow.

DesignPrototypeHTML

Layout

W358

H48

X16

Y772

Classes

w-full h-12 rounded-xl bg-gray-900 text-white font-semibold mt-auto

Fill

#111827 gray-900

Layout OK · 0 issues

Your screens keep your design’s colors: only the interface around them wears Saymock’s.

Same agent, same brief.Fewer tokens, faster.

Claude Opus 5.5 reproduced three Wikipedia screens as editable designs, with Saymock and then with Figma (MCP). Same prompt, same assets, same budget.

Tokens per screen

−31%fewer tokens

1.88M with Saymock2.74M with Figma (MCP)

Cost per screen

−37%fewer dollars

$1.70 with Saymock$2.69 with Figma (MCP)

Time per screen

1.8×faster

5.4 min with Saymock9.8 min with Figma (MCP)

Design quality

Equal qualitySaymock 79.0, Figma 82.2 out of 100: under 5 points apart, our protocol calls it a tie.

Visual fidelity /60Editability /40

3 Wikipedia screens (iPhone, Android, web) × 3 runs per tool, Claude Opus 5.5, double-blind scoring, November 12, 2026. Tokens and dollars measured by Claude Code, at API prices.

Protocol and results

The format your agent knows best

No coordinates to compute, no proprietary format: your agent reads and writes HTML with Tailwind classes. Repeated elements are written once as a template, and every element keeps its identity from one rewrite to the next.

  1. Add the server to Claude Code: **claude mcp add** --scope user --transport http saymock https://app.saymock.com/mcp Copy
  2. In Claude Code, run /mcp and sign in to Saymock. Claude (web and desktop), Cursor and VS Code: Settings → Connect your agent.
  3. Ask for a screen. To handle your comments as you review, run /saymock:session.

create_frame edit_html update_nodes get_html get_screenshot check_layout wait_for_tasks reply_comment

<div data-id="k2f9" class="flex flex-col gap-3 rounded-2xl bg-white p-6 shadow-lg">
  <h2 class="text-lg font-semibold">Pro</h2>
  <ul class="space-y-2">
    <template data-each="features">
      <li class="flex gap-2">
        <i data-lucide="check" class="size-4"></i>{{.}}
      </li>
    </template>
  </ul>
  <button data-goto="checkout" class="rounded-xl bg-zinc-900 py-3 text-white">
    Choose plan
  </button>
</div>

data-id keeps comments and selections · data-each repeats an element · data-goto links prototype screens

Frequently asked questions

Do I have to pay for an AI on top?

Saymock never calls a model itself. You plug in your own agent, such as Claude Code: it works with your account and can read your code to design screens that are true to your product.

Which agents work with it?

Any client that speaks MCP over Streamable HTTP. A personal access token links the agent to your account and your workspace.

Are my designs real code?

Yes. Every screen is HTML with Tailwind v4 classes, no scripts. Copy it into your project, or ask your agent to build it in your stack.

Can I use it without an agent?

Yes. The editor stands on its own: frames and devices, a Tailwind inspector, layers, prototypes, comments and sharing.

Who can see my files?

The members of your workspace, and the people you send a share link to, with the role you chose: view, comment or edit.